I'm not really sure. I recently found a paper (I think from Vilnius univ) where they
showed that the optical and electrical noise of LEDs is linked to their expected life.
Given that the Osram LEDs are so bright that it feels like they burn a hole into
the retina, it could well be that the 13 mA are already too much for them and they
might perform better with lower currents, noise-wise. And bright LEDs pay
with lifetime.
I also do not think that slope resistance of a Zener has anything to do with
its noise behaviour, and if it has, the relationship seems more to be inverse.
Take a look at the BZX84 plots. The low voltage ones have the lowest noise,
yet the roundest knees; the high voltage ones have a sharp knee aka low
resistance and they are LOUD.
Low resistance comes from amplification; be it explicit like in a
bandgap or by carrier multiplication like in avalanche breakdown.
And the noise is amplified, too.
I'll try to measure that in the foreseeable future, but right now the available
time goes to my multiple IF3602 preamp and its interaction with Schottky
ring mixers used for phase noise measurements.
regards, Gerhard
Last edited:
You made a proposal for a low noise amplifier here some time
ago: http://www.diyaudio.com/forums/part...rements-leds-zener-diodes-18.html#post4692727
It may be useful to organize a group buy for the circuit board.
As far as I can see this is the most versatile of all low noise
preamps seen on diyaudio?
ago: http://www.diyaudio.com/forums/part...rements-leds-zener-diodes-18.html#post4692727
It may be useful to organize a group buy for the circuit board.
As far as I can see this is the most versatile of all low noise
preamps seen on diyaudio?
I would not object it, but I also would not organize it. The Gerber data
is on my web site in the same directory. It is slightly corrected for better
available parts. That includes ceramic capacitors for the input. I could
not find any of the often cited drawbacks. But if you really want to go
down to 0.1 Hz, you need 10000uF anyway to short the noise of the bias
resistor to the low-impedance input. That may have to be a wet slug
tantal, and that costs an arm and a leg. And when connecting it with
the large cap, you need some discipline to short the op amp input side
to gnd until the capacitor is charged. The op amps might not like that.
I'm currently working on a version with paralleled IF3602, also
not really cheap, but somewhat more relaxed wrt the input cap.
regards, Gerhard
You're absolutely right. The low voltage Zeners exhibit the Zener effect, which has a soft knee and generates purely shot noise. The higher voltage 'Zeners' exhibit the avalanche effect, and as you state, this effect also includes carrier multiplication, not just a simple shot noise process. So, for commercial 'Zeners', the two effects are present in varying amounts, selectable according to their rated breakdown voltage.
Hi,
I got 220 pV/sqrt Hz from averaging 20 ADA4898 op amps. It is closed loop, accuracy
and temperature independence come from feedback and it simply behaves as calculated.
Noise performance can easily be verified by measuring the thermal noise of resistors.
The drawback is that the bias system is medium impedance and one needs a large
input coupling capacitor if one wants a lower corner of 0.1 Hz. After intensive fight with
myself I have ordered some wet slug tantalums yesterday.
(AVX seems to be cheaper than Vishay, but still OMG.)
I'm also experimenting with a flock of IF3602s, that won't be cheaper. My signal source
can live with a Cin of a nF or two, but 1/f is better with the 4898. The only real
advantage would be the polarity independence and the harmless current surges
through the small coupling capacitor. Vcc must be ultra-clean, even
a LT3042 needs some help. Maybe I can buffer the 2nd stage with a BUF634 or so
and feed the pull up from there. Should give some PSSR.
With the ADA4898, PSSR comes for free.
regards, Gerhard
Gerhard,
Parallel opamps can lower the voltage noise by sqrt(N), where N is the number of opamp.
How about the current noise? Will parallel opamps reduce it or increase it?
It increases by the same power law. This is an advantage of JFETs, as the parallel ("current") noise is so small to begin with, at least up to about 100kHz, when one begins to see induced gate noise from the channel thermal noise.
bcarso, thank you for the explanation.
I do have the same thought.
I have done some calculation before. It seems that the resulting current noise is quite large unless we can keep the resistors values very low. As far as I know, most bjt opamps' current noise lies 1.x ~ 2.x pA /sqrt Hz. After parallell, the noise voltage will be in the same magnitude as the voltage noise of the bjt opamp used. However, the input impedance cannot be so low when using bjt opamps. Right?
So, I expect that there may be limit or a sweet spot about the number of parallel opamps.
I do have the same thought.
I have done some calculation before. It seems that the resulting current noise is quite large unless we can keep the resistors values very low. As far as I know, most bjt opamps' current noise lies 1.x ~ 2.x pA /sqrt Hz. After parallell, the noise voltage will be in the same magnitude as the voltage noise of the bjt opamp used. However, the input impedance cannot be so low when using bjt opamps. Right?
So, I expect that there may be limit or a sweet spot about the number of parallel opamps.
Think in terms of the parallel noise flowing in the source impedance, not the input impedance of the op amps. This, for many interesting measurements, or for applications like moving coil cartridge preamps, can be quite low. For example suppose the source impedance is 10 ohms. Suppose our candidate op amps have voltage noise density of 1nV/sq rt Hz and parallel "current" noise (at each input) of 2pA/sq rt Hz. What is the optimal number of op amps to be paralleled? Let's suppose we use them in the noninverting configuration, and have a very low impedance feedback network to set the gain so it has a negligible contribution to total noise. So we need only consider the parallel noise at the noninverting input.
For the moment let's suppose the 10 ohm source has no thermal noise (it's in a 0 Kelvin refrigerator). With one op amp we see the 1nV/sq rt Hz and 10 ohms times 2pA/ sq rt Hz = 20pV/sq rt Hz. Clearly the 1nV dominates strongly.
Knowing that a given result is a value of a smoothly varying function except for the lumpiness of N being an integer, we can determine the lowest value of e sub n attainable from methods of differential calculus, or at least by trial-and-error. For example, parallel 100 such amps. The voltage noise is now 100pV/sq rt Hz and the parallel noise is ten times 2pA/sq rt Hz times 10 ohms, or 200pV/sq rt Hz. The total input-referred noise assuming no correlation between e sub n and i sub n will be the rms sum, or 223.61pV/ sq rt Hz. So with 100, we have gone too far. The optimal number is related to what is more frequently shown as part of the expression for optimal source resistance, given series ("voltage") noise and parallel "current" noise, as e sub n/i sub n, so in this case we can derive a quadratic equation in N, the number of paralleled amps. The real positive root of that equation will be some value for N, unlikely an integer. So we pick the closest integer. Or look at this as making the "current" noise contribution flowing in 10 ohms as optimally being the same as the voltage noise contribution.
I'm too lazy to solve directly for the roots of the quadratic right now! It has been a very long day. But one can see where this goes. And I think the argument leads to a simpler expression.
So I just guessed Nopt is 50, and that seems to check. Total noise density will be the square root of the expression ([1nV/(square root of 50)]^2 + [(square root of 50)(2pA)(10 ohms)]^2), or 200pV/sq rt Hz. A big improvement over a single amp but a lot of them to get there.
So the shortcut is: determine the optimal source impedance for noise for the candidate op amp. In this case that is 500 ohms. Divide that by the actual source impedance, in this example 10 ohms. The answer is the nearest integer to that quotient, and in this case it is exactly an integer: 50.
One can see why amplifiers with negligible parallel noise are useful, although integrated JFET op amps' series noise is usually much higher than bipolar types.
Last edited:
Really good explanation.
You can see the noise vs. source resistance parabola in every data sheet
of op amps or transistors that assume bragging rights about their noise performance.
The sweet spot is simply the lowest point.
In the data sheet, it is usually opt. source resistance for one op amp, but n op amps
for a given source impedance is the same game.
You can see the noise vs. source resistance parabola in every data sheet
of op amps or transistors that assume bragging rights about their noise performance.
The sweet spot is simply the lowest point.
In the data sheet, it is usually opt. source resistance for one op amp, but n op amps
for a given source impedance is the same game.
yes.
the trick with the parallel opamps is shown on page 17 of the LT1028 data sheet.
< http://cds.linear.com/docs/en/datasheet/1028fd.pdf >
Below a source resistance of, say, 50 Ohms one can win with multiple LT1028.
the trick with the parallel opamps is shown on page 17 of the LT1028 data sheet.
< http://cds.linear.com/docs/en/datasheet/1028fd.pdf >
Below a source resistance of, say, 50 Ohms one can win with multiple LT1028.
Attachments
Last edited:
bcarso,
Not until this morning did I really how the source impedance interact with the current noise.
I also found that Douglas Self smartly chooses 5534/5532 for parallel opamp. Because these opamps are the few bjt opamps which have current noise lower than 1pA/sqrt Hz.
Let's compare doubling AD797 (0.9nV and 2pA) and 5534 (3.5nV and 0.6pA) with Rs = 500. For doubling AD797, we get 0.64nV and 2pA x 500 x sqrt(2) = 1.41nV. The benefit of decreasing voltage noise is counteracted by the increased (and comparable) current noise. For doubling 5534, we get 2.47nV and 0.6pA x 500 x sqrt(2) = 0.42 nV. We can see that the increase in current noise is less significant due to the small base value 0.6pA.
Douglas Self wisely chooses 5534 for parallel arrangement because of its low current noise (but the voltage noise is large enough).
One may experiment with OPA1611, LME49710 and LME49990 see how current noise may hamper the effect of parallel op amps.
Not until this morning did I really how the source impedance interact with the current noise.
I also found that Douglas Self smartly chooses 5534/5532 for parallel opamp. Because these opamps are the few bjt opamps which have current noise lower than 1pA/sqrt Hz.
Let's compare doubling AD797 (0.9nV and 2pA) and 5534 (3.5nV and 0.6pA) with Rs = 500. For doubling AD797, we get 0.64nV and 2pA x 500 x sqrt(2) = 1.41nV. The benefit of decreasing voltage noise is counteracted by the increased (and comparable) current noise. For doubling 5534, we get 2.47nV and 0.6pA x 500 x sqrt(2) = 0.42 nV. We can see that the increase in current noise is less significant due to the small base value 0.6pA.
Douglas Self wisely chooses 5534 for parallel arrangement because of its low current noise (but the voltage noise is large enough).
One may experiment with OPA1611, LME49710 and LME49990 see how current noise may hamper the effect of parallel op amps.
I just go with a bunch of JFETs these days, but it can be cumbersome. Ideally they preface a decent bipolar op amp (1611 seems pretty decent and likely to be made for a while, as opposed to a bunch of formerly National parts TI is obsoleting because there is undue overlap with TI's own). In one case to (slightly) simplify things I used a BF862 as a follower prefacing a 1611, in which case the JFET noise added rms with the op amp's 1.1nV/sq rt Hz and the feedback divider noise. But this was not an all-out attempt to get super low noise, just something for an "adequate" preamp for a product.
5532/5534 are amazing values and astonishing in their longevity! Self uses them as much as he can, and in some cases misses the boat a bit (like appending one to realize a synthetic 47k termination for a phono cartridge---see Small Signal Audio Design. But overall Doug does know his stuff.
5532/5534 are amazing values and astonishing in their longevity! Self uses them as much as he can, and in some cases misses the boat a bit (like appending one to realize a synthetic 47k termination for a phono cartridge---see Small Signal Audio Design. But overall Doug does know his stuff.